Surgical Feature-Space Decomposition of LLMs: Why, When and How?
- URL: http://arxiv.org/abs/2405.13039v1
- Date: Fri, 17 May 2024 07:34:03 GMT
- Title: Surgical Feature-Space Decomposition of LLMs: Why, When and How?
- Authors: Arnav Chavan, Nahush Lele, Deepak Gupta,
- Abstract summary: We empirically study the efficacy of weight and feature space decomposition in transformer-based language models.
We show that surgical decomposition provides critical insights into the trade-off between compression and language modelling performance.
We extend our investigation to the implications of low-rank approximations on model bias.
- Score: 8.826164604720738
- License: http://creativecommons.org/licenses/by-nc-sa/4.0/
- Abstract: Low-rank approximations, of the weight and feature space can enhance the performance of deep learning models, whether in terms of improving generalization or reducing the latency of inference. However, there is no clear consensus yet on \emph{how}, \emph{when} and \emph{why} these approximations are helpful for large language models (LLMs). In this work, we empirically study the efficacy of weight and feature space decomposition in transformer-based LLMs. We demonstrate that surgical decomposition not only provides critical insights into the trade-off between compression and language modelling performance, but also sometimes enhances commonsense reasoning performance of LLMs. Our empirical analysis identifies specific network segments that intrinsically exhibit a low-rank structure. Furthermore, we extend our investigation to the implications of low-rank approximations on model bias. Overall, our findings offer a novel perspective on optimizing LLMs, presenting the low-rank approximation not only as a tool for performance enhancements, but also as a means to potentially rectify biases within these models. Our code is available at \href{https://github.com/nyunAI/SFSD-LLM}{GitHub}.
Related papers
- Model Surgery: Modulating LLM's Behavior Via Simple Parameter Editing [63.20133320524577]
Large Language Models (LLMs) have demonstrated great potential as generalist assistants.
It is crucial that these models exhibit desirable behavioral traits, such as non-toxicity and resilience against jailbreak attempts.
In this paper, we observe that directly editing a small subset of parameters can effectively modulate specific behaviors of LLMs.
arXiv Detail & Related papers (2024-07-11T17:52:03Z) - Quantifying Emergence in Large Language Models [31.608080868988825]
We propose a quantifiable solution for estimating emergence of LLMs.
Inspired by emergentism in dynamics, we quantify the strength of emergence by comparing the entropy reduction of the macroscopic (semantic) level with that of the microscopic (token) level.
Our method demonstrates consistent behaviors across a suite of LMs under both in-context learning (ICL) and natural sentences.
arXiv Detail & Related papers (2024-05-21T09:12:20Z) - What Makes Quantization for Large Language Models Hard? An Empirical
Study from the Lens of Perturbation [55.153595212571375]
Quantization is a technique for improving the memory and computational efficiency of large language models (LLMs)
We propose a new perspective on quantization, viewing it as perturbations added to the weights and activations of LLMs.
We conduct experiments with various artificial perturbations to explore their impact on LLM performance.
arXiv Detail & Related papers (2024-03-11T03:42:51Z) - Characterizing Truthfulness in Large Language Model Generations with
Local Intrinsic Dimension [63.330262740414646]
We study how to characterize and predict the truthfulness of texts generated from large language models (LLMs)
We suggest investigating internal activations and quantifying LLM's truthfulness using the local intrinsic dimension (LID) of model activations.
arXiv Detail & Related papers (2024-02-28T04:56:21Z) - In-context Learning and Gradient Descent Revisited [3.085927389171139]
We show that even untrained models achieve comparable ICL-GD similarity scores despite not exhibiting ICL.
Next, we explore a major discrepancy in the flow of information throughout the model between ICL and GD, which we term Layer Causality.
We propose a simple GD-based optimization procedure that respects layer causality, and show it improves similarity scores significantly.
arXiv Detail & Related papers (2023-11-13T21:42:38Z) - CRaSh: Clustering, Removing, and Sharing Enhance Fine-tuning without
Full Large Language Model [22.870512676002463]
This paper focuses on Offsite-Tuning (OFT), a representative technique that transfers transformer blocks between centralized LLMs and downstream emulators.
Inspired by these observations, we propose CRaSh, involving Clustering, Removing, and Sharing, a training-free strategy to derive improved emulators from LLMs.
Our findings demonstrate a linear connectivity among these optima falling over the same basin, thereby highlighting the effectiveness of CRaSh and OFT.
arXiv Detail & Related papers (2023-10-24T03:08:58Z) - On the Relation between Internal Language Model and Sequence Discriminative Training for Neural Transducers [52.88268942796418]
Internal language model (ILM) subtraction has been widely applied to improve the performance of the RNN-Transducer.
We show that sequence discriminative training has a strong correlation with ILM subtraction from both theoretical and empirical points of view.
arXiv Detail & Related papers (2023-09-25T13:35:28Z) - Explaining Emergent In-Context Learning as Kernel Regression [61.57151500616111]
Large language models (LLMs) have initiated a paradigm shift in transfer learning.
In this paper, we investigate the reason why a transformer-based language model can accomplish in-context learning after pre-training.
We find that during ICL, the attention and hidden features in LLMs match the behaviors of a kernel regression.
arXiv Detail & Related papers (2023-05-22T06:45:02Z) - R\'{e}nyi Divergence Deep Mutual Learning [10.357597530261664]
This paper revisits Deep Learning Mutual (DML) as a simple yet effective computing paradigm.
We propose using R'enyi divergence instead of the KL divergence, which is more flexible and limited.
Our empirical results demonstrate the advantage combining DML and R'enyi divergence, leading to further improvement in model generalization.
arXiv Detail & Related papers (2022-09-13T04:58:35Z) - Toward Fast, Flexible, and Robust Low-Light Image Enhancement [87.27326390675155]
We develop a new Self-Calibrated Illumination (SCI) learning framework for fast, flexible, and robust brightening images in real-world low-light scenarios.
Considering the computational burden of the cascaded pattern, we construct the self-calibrated module which realizes the convergence between results of each stage.
We make comprehensive explorations to SCI's inherent properties including operation-insensitive adaptability and model-irrelevant generality.
arXiv Detail & Related papers (2022-04-21T14:40:32Z) - Locally Interpretable Model Agnostic Explanations using Gaussian
Processes [2.9189409618561966]
Local Interpretable Model-Agnostic Explanations (LIME) is a popular technique for explaining the prediction of a single instance.
We propose a Gaussian Process (GP) based variation of locally interpretable models.
We demonstrate that the proposed technique is able to generate faithful explanations using much fewer samples as compared to LIME.
arXiv Detail & Related papers (2021-08-16T05:49:01Z)
This list is automatically generated from the titles and abstracts of the papers in this site.
This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.